Will this relationship actually make it to a marriage that works?
Thirty honest answers about the two of you, your families, money and timing. The tool weighs them the way the research does, and tells you which few things are deciding the outcome.
- Couples behind the weights
- 11,196Pooled across 43 longitudinal datasets
- Questions
- 30Eight steps, about four minutes
- Stages modelled
- TwoReaching marriage × staying happy in it
- Data sent anywhere
- NoneUnless you choose to share it
What it asks about
This is a structured gut-check, not a verdict. The weights are informed priors drawn from published effect sizes, not coefficients fitted to Indian data — no public dataset follows Indian dating couples into marriage yet. Read Limits before you take any number here seriously. If there is violence, control or coercion in your relationship, the tool will stop scoring and say so instead.
Every answer, in one place
Overall = Stage 1 × Stage 2. Both have to go right.
Pulling it down most
Holding it up most
Net pull by area
Everything above is correlation, not destiny. Every single item in the list is something couples change — treat Pulling it down most as a to-do list, and have your partner fill this in separately. A big gap between your two sets of answers is itself one of the clearest signals in the research.
How the score is built
Nothing here is a black box. Every weight sits in one readable file, and this page explains where each one came from.
The arithmetic
Every answer becomes a number between −1 and +1 (the middle of a slider is 0). Each one is multiplied by a weight and added up — separately for two stages: getting married and staying in a good marriage. Each total goes through a logistic curve, squeezed into 3% to 95%. The two stage probabilities are then multiplied.
overall = pstage 1 × pstage 2
Why cap at 95% and floor at 3%? People change, jobs move, parents fall ill. No questionnaire can rule those out, so the tool never claims near-certainty in either direction. The likely range under the headline number shows what happens if your answers are a bit too rosy or a bit too gloomy — it shifts both stage totals by ±1 and re-runs the arithmetic.
Weights are ranked by how strong and how well-replicated the evidence is. The best-tested predictors — commitment, contempt, trust, money conflict, family approval — carry the most. Individual traits like personality get none, on purpose: the largest pooled study found they explain far less than how the two people feel about the relationship itself.
Evidence behind each domain
| Domain | Max pull | What the research says | India angle |
|---|---|---|---|
| The two of youcommitment, respect, conflict, closeness | high | A machine-learning study pooling 11,196 couples from 43 longitudinal datasets found the top predictors of relationship quality were perceived partner commitment, appreciation, sexual satisfaction, perceived partner satisfaction and conflict. Gottman's lab work names contempt as the single worst sign; his headline accuracy figures are in-sample, so treat them as direction, not precision. Stanley & Rhoades found that ever seeing your partner as less committed than you, before marriage, predicted unhappiness after. | Applies across cultures. Pressure to “adjust” can hide contempt, so the tool asks about behaviour, not feelings alone. |
| Families | high (stage 1) | The “Romeo and Juliet effect” — opposition making love stronger — mostly failed to replicate. The opposite, the social network effect, holds up: approval from family and friends goes with more love, commitment and stability. | Pew's 2021 survey of ~30,000 Indians: 64% say it is very important to stop women in their community marrying into another caste, and 67% of Hindus say the same about inter-faith marriage. A national youth survey found 84% of married youth had parent-arranged marriages and only 4% were inter-caste. A cross-community couple faces a real structural barrier, and the tool prices that in as a family hurdle. It is not a judgement on the couple. |
| Other people in the picture | very high | Rusbult's investment model: commitment = satisfaction + investment − attractive alternatives. A meta-analysis of 52 studies (~11,000 people, 5 countries) confirmed all three predict commitment and breakup. An active outside romance or a hidden close bond is the “alternatives” term at its worst; secrecy also erodes trust directly. | A family-pushed rishta counts as an alternative too. That is why “pressure to marry someone else” sits in this part of the model. |
| Money & careers | medium–high | Across 4,574 US couples, money arguments predicted divorce more strongly than fights about kids, sex or in-laws, even after controlling for income. The arguments did the damage, not the bank balance. | Wedding cost, dowry-style demands (illegal under the Dowry Prohibition Act, 1961), whether the wife keeps working, and supporting parents are common flashpoints here. |
| Age & timeline | low–medium | US data shows divorce risk falls through the early-to-mid twenties and flattens after. Very short relationships give little information; very long ones with no plan point to “sliding” rather than deciding. | Legal minimum: 21 for men, 18 for women. Family pressure tends to spike for women in their late twenties, which can force a timeline clash. |
| Life plans & expectations | medium | Mismatched expectations about roles, kids and religion show up later as recurring conflict, which feeds back into the conflict and money terms above. | Joint versus nuclear family, in-law involvement and gender roles weigh heavier here than in most Western samples. |
| Effort & communication | medium | Couples who make an explicit decision to commit — “deciding, not sliding” — do better than those who drift. Repair after fights is one of Gottman's strongest protective signs. | Includes having a concrete plan to bring both families round, which in practice decides a lot of Indian love marriages. |
| Safety check | override | Violence, threats and control are not risk factors to be traded off against good scores elsewhere. If any is ticked, the tool shows a safety notice instead of a reassuring number. | Women Helpline: 181. Emergency: 112. |
Sources
- Joel et al., PNAS 2020 — machine-learning predictors of relationship quality across 43 longitudinal studies
- The Gottman Institute — divorce prediction and the role of contempt
- Stanley, Institute for Family Studies — asymmetric commitment before marriage
- Sinclair et al. — the social network effect versus the Romeo and Juliet effect
- Felmlee & Sprecher — longitudinal study on network approval
- Pew Research Center, 2021 — Religion in India: Tolerance and Segregation
- Indian youth survey on marriage choice — 84% arranged, 4% inter-caste
- Dew, Britt & Huston, 2012 — financial issues and divorce
- Le & Agnew, 2003 — meta-analysis of the investment model, Personal Relationships 10:37–57
What this can and can't tell you
Read this before you let a number change a decision. The honest summary: it is a structured way to think, not a prediction.
Five real limits
- It isn't fitted to Indian data. No public longitudinal dataset follows Indian dating couples into marriage and beyond. The weights are informed priors from published effect sizes, not regression coefficients.
- It's one person's view. Self-reports lean optimistic, and you only know your side. Have your partner fill it in separately and compare. Big gaps between your two sets of answers are themselves a signal.
- Most source studies are Western. The relational predictors — commitment, contempt, trust — replicate widely. Family and community effects are much stronger in India, so those carry extra weight here.
- Correlation, not destiny. Every factor here is something couples change. Use the “pulling it down” list as a to-do list.
- “Success” means married and reasonably happy. India's low formal divorce rate partly reflects stigma, so “still married” alone is a poor success measure. Stage 2 is about quality, not just staying together.
Your data
Nothing you type leaves your browser unless you press Share anonymously on the result page. Answers are kept in this browser's local storage so you can come back to them, and Start blank clears them.
How to turn this into a real model
- Collect a panel: both partners answer these items separately at baseline, then follow up at 12, 24 and 60 months — married? satisfaction score?
- Fit two logistic regressions (married by 24 months; satisfied at 60 months) with a hierarchical prior centred on the current weights, so small samples shrink toward the literature.
- Check calibration — reliability curve, Brier score — and AUC on held-out couples. Report by region and community, since effects differ a lot.
- Replace the fixed 3–95% squeeze with a posterior interval from the fitted model.
Report back
Shared your answers earlier? Come back in a year or two and tell us what happened — this is the part that lets the model actually learn. Use the code you were given when you shared.